Classification of river ice using polarimetric SAR data

Ice jams are a major cause of river flooding in Canada. These events can be devastating for the environment, human infrastructure, and population. Although methodologies have been developed to discriminate ice types using single-polarization synthetic aperture radar (SAR) data, SAR polarimetry has not yet been used. In this paper a polarimetric SAR airborne image of the Saint-François River, Quebec, has been analyzed. Complementary data about the characteristics of the ice cover were obtained simultaneously with the image acquisition. The usefulness of each polarimetric parameter is explored to obtain realistic ice type classifications. We propose to compute a rule-based hierarchical classification and compare it with a Wishart classification. A single-polarization-based classification is also used to show the limits of this approach in discriminating water from ice. The hierarchical classification more accurately separates areas of ice from areas of open water (81% producer’s accuracy). Both classifications show good results, with few ambiguities in detection of the consolidated ice class. Detection of the thermal ice class is not highly accurate. Thermal and frazil ice classification is performed better when hierarchical classification than when Wishart classification is used. Lastly, the hierarchical classification is better adapted to river ice than Wishart classification, and fully polarimetric data are significantly better than single-polarization data for discriminating water from ice.